1. CS 440 / ECE 448 Introduction to Artificial Intelligence Spring 2010 Lecture #3 Instructor: Eyal Amir Grad TAs: Wen Pu, Yonatan Bisk Undergrad TAs: Sam Johnson, Nikhil Johri

2. To DFS / BFS – Deprived If you do not know what DFS or BFS are, and Ch. 3 is not enough of a review for you, please see the TAs ASAP –CS173 (Prereq.) covers BFS,DFS in detail –Alternative classes in ECE should have covered those (let me know if not)

3. Problem-Solving Agent environment agent ? sensors actuators

4. Problem-Solving Agent environment agent ? sensors actuators Formulate Goal Formulate Problem States Actions Find Solution

5. Example Problem Start Street Street with Parking

6. Looking for Parking Going home; need to find street parking Formulate Goal: Car is parked Formulate Problem: States: street with parking and car at that street Actions: drive between street segments Find solution: Sequence of street segments, ending with a street with parking

7. Problem Formulation Start Street Street with Parking Search Path

8. Search Problem State space –each state is an abstract representation of the environment –the state space is discrete Initial state Successor function Goal test Path cost

9. Search Problem State space Initial state: –usually the current state –sometimes one or several hypothetical states (“what if …”) Successor function Goal test Path cost

10. Search Problem State space Initial state Successor function: –[state  subset of states] –an abstract representation of the possible actions Goal test Path cost

11. Search Problem State space Initial state Successor function Goal test: –usually a condition –sometimes the description of a state Path cost

12. Search Problem State space Initial state Successor function Goal test Path cost: –[path  positive number] –usually, path cost = sum of step costs –e.g., number of moves of the empty tile

13. Assumptions in Basic Search The environment is static The environment is discretizable The environment is observable The actions are deterministic

14. Search Space Size Unlike Search in CS225, AI encounters search spaces that are too large AI Search typically does not realize the entire search graph or state space Examples –Scheduling CS classes such that every student in every program of study can take every class they wish –Search for shortest path that covers all streets (Travelling Salesman Problem)

15. Search Space Size Scheduling CS classes such that every student in every program of study can take every class they wish States = ? State Space Size = ? Search Time = ?

16. Search Space Size Search for shortest path that covers all streets (Travelling Salesman Problem) State = ? State Space Size = ? Search Time = ?

17. Simple Agent Algorithm Problem-Solving-Agent 1.initial-state  sense/read state 2.goal  select/read goal 3.successor  select/read action models 4.problem  (initial-state, goal, successor) 5.solution  search(problem) 6.perform(solution)

18. Basic Search Concepts Search tree Search node Node expansion Search strategy: At each stage it determines which node to expand

19. Node Data Structure STATE PARENT ACTION COST DEPTH If a state is too large, it may be preferable to only represent the initial state and (re-)generate the other states when needed If a state is too large, it may be preferable to only represent the initial state and (re-)generate the other states when needed

20. Fringe Set of search nodes that have not been expanded yet Implemented as a queue FRINGE –INSERT(node,FRINGE) –REMOVE(FRINGE) The ordering of the nodes in FRINGE defines the search strategy

21. Search Algorithm 1.If GOAL?(initial-state) then return initial-state 2.INSERT(initial-node,FRINGE) 3.Repeat: If FRINGE is empty then return failure n  REMOVE(FRINGE) s  STATE(n) For every state s’ in SUCCESSORS(s)  Create a node n’  If GOAL?(s’) then return path or goal state  INSERT(n’,FRINGE)

22. Search Strategies A strategy is defined by picking the order of node expansion Performance Measures: –Completeness – does it always find a solution if one exists? –Time complexity – number of nodes generated/expanded –Space complexity – maximum number of nodes in memory –Optimality – does it always find a least-cost solution Time and space complexity are measured in terms of –b – maximum branching factor of the search tree –d – depth of the least-cost solution –m – maximum depth of the state space (may be ∞)

23. Remark Some problems formulated as search problems are NP-hard problems. We cannot expect to solve such a problem in less than exponential time in the worst-case But we can nevertheless strive to solve as many instances of the problem as possible

24. Blind vs. Heuristic Strategies Blind (or uninformed) strategies do not exploit any of the information contained in a state Heuristic (or informed) strategies exploits such information to assess that one node is “more promising” than another

25. Blind Strategies Breadth-first –Bidirectional Depth-first –Depth-limited –Iterative deepening Uniform-Cost Step cost = 1 Step cost = c(action)   > 0

26. Breadth-First Strategy New nodes are inserted at the end of FRINGE 23 45 1 67 FRINGE = (1)

27. Breadth-First Strategy New nodes are inserted at the end of FRINGE FRINGE = (2, 3) 23 45 1 67

28. Breadth-First Strategy New nodes are inserted at the end of FRINGE FRINGE = (3, 4, 5) 23 45 1 67

29. Breadth-First Strategy New nodes are inserted at the end of FRINGE FRINGE = (4, 5, 6, 7) 23 45 1 67

30. Evaluation b: branching factor d: depth of shallowest goal node Complete Optimal if step cost is 1 Number of nodes generated: 1 + b + b 2 + … + b d + b(b d -1) = O(b d+1 ) Time and space complexity is O(b d+1 )

31. Time and Memory Requirements d#NodesTimeMemory 2111.01 msec11 Kbytes 411,1111 msec1 Mbyte 6~10 6 1 sec100 Mb 8~10 8 100 sec10 Gbytes 10~10 10 2.8 hours1 Tbyte 12~10 12 11.6 days100 Tbytes 14~10 14 3.2 years10,000 Tb Assumptions: b = 10; 1,000,000 nodes/sec; 100bytes/node

32. Time and Memory Requirements d#NodesTimeMemory 2111.01 msec11 Kbytes 411,1111 msec1 Mbyte 6~10 6 1 sec100 Mb 8~10 8 100 sec10 Gbytes 10~10 10 2.8 hours1 Tbyte 12~10 12 11.6 days100 Tbytes 14~10 14 3.2 years10,000 Tb Assumptions: b = 10; 1,000,000 nodes/sec; 100bytes/node

33. Bidirectional Strategy 2 fringe queues: FRINGE1 and FRINGE2 Time and space complexity = O(b d/2 ) << O(b d )

34. Depth-First Strategy New nodes are inserted at the front of FRINGE 1 23 45 FRINGE = (1)

35. Depth-First Strategy New nodes are inserted at the front of FRINGE 1 23 45 FRINGE = (2, 3)

36. Depth-First Strategy New nodes are inserted at the front of FRINGE 1 23 45 FRINGE = (4, 5, 3)

37. Depth-First Strategy New nodes are inserted at the front of FRINGE 1 23 45

38. Depth-First Strategy New nodes are inserted at the front of FRINGE 1 23 45

39. Depth-First Strategy New nodes are inserted at the front of FRINGE 1 23 45

40. Depth-First Strategy New nodes are inserted at the front of FRINGE 1 23 45

41. Depth-First Strategy New nodes are inserted at the front of FRINGE 1 23 45

42. Depth-First Strategy New nodes are inserted at the front of FRINGE 1 23 45

43. Depth-First Strategy New nodes are inserted at the front of FRINGE 1 23 45

44. Depth-First Strategy New nodes are inserted at the front of FRINGE 1 23 45

45. Evaluation b: branching factor d: depth of shallowest goal node m: maximal depth of a leaf node Complete only for finite search tree Not optimal Number of nodes generated: 1 + b + b 2 + … + b m = O(b m ) Time complexity is O(b m ) Space complexity is O(bm)

46. Depth-Limited Strategy Depth-first with depth cutoff k (maximal depth below which nodes are not expanded) Three possible outcomes: –Solution –Failure (no solution) –Cutoff (no solution within cutoff)

47. Iterative Deepening Strategy Repeat for k = 0, 1, 2, …: Perform depth-first with depth cutoff k Complete Optimal if step cost =1 Time complexity is: (d+1)(1) + db + (d-1)b 2 + … + (1) b d = O(b d ) Space complexity is: O(bd)

48. Comparison of Strategies Breadth-first is complete and optimal, but has high space complexity Depth-first is space efficient, but neither complete nor optimal Iterative deepening is asymptotically optimal

49. Repeated States 8-queens No assembly planning Few 123 45 678 8-puzzle and robot navigation Many search tree is finitesearch tree is infinite

50. Avoiding Repeated States Requires comparing state descriptions Breadth-first strategy: –Keep track of all generated states –If the state of a new node already exists, then discard the node

51. Avoiding Repeated States Depth-first strategy: –Solution 1: Keep track of all states associated with nodes in current tree If the state of a new node already exists, then discard the node  Avoids loops –Solution 2: Keep track of all states generated so far If the state of a new node has already been generated, then discard the node  Space complexity of breadth-first

52. Detecting Identical States Use explicit representation of state space Use hash-code or similar representation

53. Uniform-Cost Strategy Each step has some cost   > 0. The cost of the path to each fringe node N is g(N) =  costs of all steps. The goal is to generate a solution path of minimal cost. The queue FRINGE is sorted in increasing cost. S 0 1 A 5 B 15 C SG A B C 5 1 10 5 5 G 11 G 10

54. Modified Search Algorithm 1.INSERT(initial-node,FRINGE) 2.Repeat: If FRINGE is empty then return failure n  REMOVE(FRINGE) s  STATE(n) If GOAL?(s) then return path or goal state For every state s’ in SUCCESSORS(s)  Create a node n’  INSERT(n’,FRINGE)

55. Branch and Bound If search involves optimization (e.g. shortest path for covering all streets) –Can record the best path so far, and bound the search when the current path becomes longer than current best path.

56. Summary Search tree  state space Search strategies: breadth-first, depth- first, and variants Evaluation of strategies: completeness, optimality, time and space complexity Avoiding repeated states Optimal search with variable step costs